codex desktop app OpenAI is crazy i have a session with nearly 300 subagents running more than a day thanks to Sisyphus Labs lazycodex and it is just smooth as hell no bad experience

imagine team mode but in your codex desktop app yes you can have this with lazycodex will be updated today stay tuned

@OpenAI so those 300 subagents were possible through lazycodex right? bro you made trully something

ok this post led my account to be suspended cuz got so viral but yo guys i am here just got back lets fuck!!!!

@OpenAI @justsisyphus why would you want to do that

@OpenAI @justsisyphus I..don’t think so? Codex has a 6 simultaneous agent cap

@OpenAI @justsisyphus running hundreds of agents that long usually leaks memory, a quick auto-restart script keeps it stable.

@OpenAI @justsisyphus running three hundred subagents for a day sounds wild, but i would worry about memory leaks or silent context drift. a smooth ui usually just masks resource contention happening under the hood.

@OpenAI @justsisyphus How much did it cost you?

@OpenAI @justsisyphus Must be nice to have a Mac and a functioning Codex app I'll just be over here in WSL tryna beat Windows

@OpenAI @justsisyphus What’s the token spend here? I’m Adding you to my eat-the-rich list.

@OpenAI @justsisyphus Breaking:

@OpenAI @justsisyphus Electron stills shine

@OpenAI @justsisyphus they're hiring 12 people a day and shipping like this. the execution is matching the headcount

@OpenAI @justsisyphus raw power!! :D

@OpenAI @justsisyphus I'm so happy for you that writing lazycodex gave you and your work more exposure

@OpenAI @justsisyphus How does the ultrawork loop compare to goal?

@OpenAI @justsisyphus it has 'subgoals' that needs to be satisfied to move on to the next goal and each subgoals requires 'evidences' after implementation so basically lazycodex not only makes implementations to be better but also does test and QA every time imagine. computer use QA is just amazing

@OpenAI @justsisyphus I mean codex sort of does both already. Is it basically just better prompts for agents or subagents or something more?

bro i think if you install this and type "ulw" with your prompt you will see those differences immediately lots of great systems like $init-deep that creates agents.md recursively programming skills that enforces type safe and memory safe code lsp tools, codegraphs, tons of great techniques are just embedded along with "teammode" in codex tools

@q_yeon_gyu_kim @OpenAI I thought team mode was opencode only?

@justsisyphus @OpenAI today codex teammode will be released

@justsisyphus @OpenAI I know this is controversial - but did you guys bring back anthropic sub use support?

@OpenAI @justsisyphus codex kills my m1 max with long threads

@OpenAI @justsisyphus And you burn through all your weeks tokens in 2 days?

@OpenAI @justsisyphus the god tibo gave me reset coupon so i am fine

@OpenAI @justsisyphus I keep considering going for $200 plan with Codex but worry about burning it out too quickly… so I keep diversifying across multiple providers 😅

@OpenAI @justsisyphus then go with that omo you will be able to leverage your kimi and glm through it

@OpenAI @justsisyphus Already have been running that stack for a couple months or so. My issue is I burn through cache with the multi agents in OmO, so I keep considering going to one provider. I kept leaning towards Codex but then Fable 5 was really good…

@q_yeon_gyu_kim @OpenAI @justsisyphus Has omo fixed the cache issue? @q_yeon_gyu_kim This seems to be a very important issue regarding how to use multi-LLMs at a time when MoA is becoming increasingly important.

@Cyberlane @OpenAI @justsisyphus we dont touch anything regarding cache

@OpenAI @justsisyphus 300 agents and ur laptop isnt a space heater rn?? thats the real flex tbh 💀

@OpenAI @justsisyphus The jump from single-agent workflows to hundreds of parallel subagents feels like a different era of software.

@OpenAI @justsisyphus I’m curious how many plans you bought, or how you save tokens.

@OpenAI @justsisyphus simple. i dont save money.

@OpenAI @justsisyphus jealous

@OpenAI @justsisyphus 300 subagents running for 24+ hours with no issues is a wild glimpse of where AI development is heading.

@OpenAI @justsisyphus competitive slopmaxxing

@OpenAI @justsisyphus Most think 300 subagents running smoothly is peak leverage. In reality, it's just automated technical debt.

@OpenAI @justsisyphus Then you’re probably the best Codex user I know. I only have a few threads at most.

@OpenAI @justsisyphus if none of them doomscrolling any social media - still can't replace me

Bro, could you share a guide how to run these sub agents in large numbers

@OpenAI @justsisyphus 평범한사업가 보고왔어용 감사합니다

@OpenAI @justsisyphus I use subagents a lot, but are you programatically creating the agents with given prompts? I ask because to me it just splits tasks accross 2-4 agents… how do you get it to orchestrate a more functional team that can actually collaborate as necessary and aligned in between?

@OpenAI @justsisyphus are you able to create one for ZCode and for Grok Build CLI ?

@OpenAI @justsisyphus 300 subagents running for a full day smoothly? That's impressive as hell. How are you handling credentials n auth across that many? With swarms that big the risk of leaks feels real.

@OpenAI @justsisyphus That's crazy—what's the maximum number of agents you've ever run at one time?

@OpenAI @justsisyphus 300 subagents running for a DAY straight and just vibing?? bro thats insane